Reliable Provision of Ancillary Services from Aggregated Variable Renewable Energy Sources through Forecasting of Extreme Quantiles

Virtual power plants aggregating multiple renewable energy sources such as Photovoltaics and Wind are promising candidates for the provision of balancing ancillary services. A requisite these services is that forecasts aggregated production need to be highly reliable in order minimize risk not providing service. Yet, a reliability greater than 99% unattainable standard forecasting models. This work proposes alternative models day-ahead prediction lowest quantiles (0.1% 0.9 %) plant production. The proposed approaches derive conditional quantile Wind/PV/Hydro production, obtained from tailored parametric machine learning models, including Convolutional Neural Network architecture predicting extremes. Reliability deviation reduced up 50 % probabilistic skill score 18% compared Quantile Regression Forest. Forecasting subsequently applied downward reserve capacity by plant. Increased leads higher capacity, but reduces average volume offered aggregation.